Deep Learning Intrusion Detection Model Based on Optimized Imbalanced Network Data

Abstract

To solve the problem of the low detection rate of minority samples in imbalanced datasets in network intrusion detection, a deep learning intrusion detection model based on optimized imbalanced data is proposed. Firstly, a hybrid sampling method is adopted in data processing. Synthetic Minority Over-sampling Technique (SMOTE) was used to increase the numbers of samples in minority categories and the majority of the samples were under-sampled by Neighborhood Cleaning Rule (NCL). Secondly, on the preprocessed balanced dataset, the high-dimensional data was reduced by Deep Belief Network (DBN) to obtain the lower low-dimensional representation of the preprocessed data. Finally, the classification work was completed by Probabilistic Neural Network (PNN). The experiment on NSL-KDD dataset showed that hybrid sampling can improve the detection rate and classification accuracy of minority categories. And the performance of DBN-PNN is obviously superior to the traditional method.